There's a lot more to optimizing comouting than just the size and number of transistors. Such as branch prediction and many other aspects. Also, gaming and ML

There's a lot more to optimizing comouting than just the size and number of transistors. Such as branch prediction and many other aspects. Also, gaming and ML training rely on massive parallelism. So that's another paradigm. More new paradigms like memory-centric and/or analog computing for AI will be commercialized in the next several years.